Episodes · 5
Every machine so far.
EP 05Out now
How a context window fills up, one token at a time
A 4,096-slot row fills token by token, each slot leaving a note on the desk; when the chat no longer fits, the amber fact (port 8123) drops off the front of the row
Agentic engineeringOct 22
EP 04Out now
Ollama vs LM Studio: what is different under the hood
The same prompt traced through both apps, side by side
How it's builtOct 5
EP 03Out now
How much VRAM you need to run an LLM locally
One token's KV-cache notes stacking into a tower that dwarfs the 2 GB model and overflows the card
How it's builtOct 4
EP 02Out now
How an LLM generates one token
Probability bars collapsing into a single token
How it's builtOct 2
EP 01Out now
How a git push reaches production
A commit traveling through CI stages to live servers
How it's builtOct 1
